arxiv: 2605.12362 · v1 · submitted 2026-05-12 · 💻 cs.NE · cs.AI

Recognition: 2 theorem links

· Lean Theorem

A Family of Quaternion-Valued Differential Evolution Algorithms for Numerical Function Optimization

\'Alvaro Gallardo, Carlos Ignacio Hern\'andez Castellanos, Gerardo Altamirano-Gomez

Authors on Pith no claims yet

Pith reviewed 2026-05-13 03:30 UTC · model grok-4.3

classification 💻 cs.NE cs.AI

keywords quaternion differential evolutionnumerical optimizationBBOB benchmarkmutation strategiesevolutionary algorithmsquaternion algebracontinuous function optimization

0 comments

The pith

Quaternion-valued differential evolution variants converge faster and perform better than standard real-valued DE on BBOB benchmark functions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a family of Quaternion-Valued Differential Evolution algorithms that run directly in quaternion space instead of real numbers. It proposes new mutation strategies built to use the algebraic and geometric traits of quaternions during the search. A reader would care because continuous function optimization appears in mechanical design, AI model training, and many engineering tasks, so any reliable speed or quality gain matters. The reported experiments show these QDE versions reach better solutions more quickly than classic DE across several function classes in the BBOB test suite.

Core claim

By moving Differential Evolution into quaternion space and designing mutation operators that exploit quaternion algebra and geometry, the new QDE algorithms produce faster convergence and higher-quality solutions than the conventional real-valued DE on multiple classes of functions in the BBOB benchmark.

What carries the argument

Quaternion mutation strategies that generate trial vectors while preserving the original DE selection and replacement rules.

If this is right

Optimization tasks in mechanical design can reach acceptable solutions in fewer evaluations.
Training routines for models that already use quaternion representations may run more efficiently.
The same quaternion extension approach could be applied to other population-based search methods.
High-dimensional continuous problems may become tractable with less computational effort.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar number-system extensions might improve performance in other evolutionary or swarm algorithms that currently use only real vectors.
The method could be checked on application-specific optimization problems from engineering rather than only on synthetic benchmarks.
If the gains hold, one could test whether further hypercomplex number systems yield additional improvements or hit diminishing returns.

Load-bearing premise

The algebraic and geometric features of quaternions can be turned into concrete search advantages by the chosen mutation rules without creating numerical problems or weakening the algorithm's reliability.

What would settle it

Executing the proposed QDE variants against standard DE on the complete BBOB suite and finding no consistent reduction in convergence time or improvement in final accuracy on the claimed function classes would disprove the performance claim.

Figures

Figures reproduced from arXiv: 2605.12362 by \'Alvaro Gallardo, Carlos Ignacio Hern\'andez Castellanos, Gerardo Altamirano-Gomez.

**Figure 3.** Figure 3: Critical difference diagram showing the ranking of t [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 4.** Figure 4: Critical difference diagram showing the ranking of t [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Critical difference diagram showing the ranking of t [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

read the original abstract

The numerical optimization of continuous functions is a fundamental task in many scientific and engineering domains, ranging from mechanical design to training of artificial intelligence models. Among the most effective and widely used algorithms for this purpose is Differential Evolution (DE), known for its simplicity and strong performance. Recent research has shown that adapting AI models to operate over alternative number systems-such as complex numbers, quaternions, and geometric algebras-can improve model compactness and accuracy. However, such extensions remain underexplored in bio-inspired optimization algorithms. In particular, the use of quaternion algebra represents an emerging area in computational intelligence. This paper introduces a family of novel Quaternion-Valued Differential Evolution (QDE) algorithms that operate directly in the quaternion space. We propose several mutation strategies specifically designed to exploit the algebraic and geometric properties of quaternions. Results show that our QDE variants achieve faster convergence and superior performance on several function classes in the BBOB benchmark compared to the traditional real-valued DE algorithm.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces new quaternion mutation strategies for differential evolution and claims better BBOB results than real-valued DE, but the abstract supplies no numbers or tests to back that up.

read the letter

The main thing to know is that the authors have built a family of quaternion-valued differential evolution algorithms with several custom mutation operators designed around quaternion multiplication and geometry. This is not a trivial swap of number systems, since non-commutativity forces real changes to how differences and scaling are handled in the DE update rule. They test on the BBOB suite and state that some variants converge faster and reach better values on certain function classes than standard real DE. That extension looks like the actual novelty, and it sits on top of existing DE literature without obvious circularity. The motivation section does a clean job explaining why quaternions have helped other models and why the same move might matter for optimization. The algebraic definitions appear consistent on first read, with no immediate contradictions in how they preserve the DE structure. The soft spot is the evidence. The abstract asserts superiority without any tables, run counts, p-values, or error bars, so the claim cannot be checked yet. If the full paper has those details and shows the gains are stable across seeds, the work strengthens; otherwise the performance story stays unverified. Minor additional questions are whether the quaternion-to-real mapping for the benchmark functions introduces extra cost or bias, and whether they compared against other non-real extensions like complex DE. This paper is mainly useful for researchers already working on evolutionary methods in domains that use quaternions, such as 3D modeling or certain control problems. A reader focused on general-purpose black-box optimization will get limited value unless the reported gains turn out to be both large and robust. It deserves a serious referee because the construction is coherent, the benchmark is standard, and the idea is new enough to warrant external checks on the experiments and comparisons. I would send it out for review rather than desk reject.

Referee Report

1 major / 1 minor

Summary. The manuscript introduces a family of Quaternion-Valued Differential Evolution (QDE) algorithms that operate directly in quaternion space. It proposes several mutation strategies designed to exploit the algebraic and geometric properties of quaternions and claims that the resulting variants achieve faster convergence and superior performance on several function classes from the BBOB benchmark relative to standard real-valued DE.

Significance. If the performance claims are substantiated with rigorous experiments, the work would be significant as one of the first systematic extensions of DE to hypercomplex numbers. It could stimulate further research on number-system adaptations in evolutionary computation, particularly for problems with natural quaternion structure such as 3-D rotations or certain signal-processing tasks. The choice of the BBOB benchmark is appropriate for testing such claims.

major comments (1)

[Abstract] Abstract: the central claim that 'our QDE variants achieve faster convergence and superior performance on several function classes in the BBOB benchmark' is asserted without any quantitative results, tables, statistical tests, error bars, or experimental-setup details. Because this empirical superiority is the load-bearing conclusion of the paper, the absence of supporting data prevents verification of the claim.

minor comments (1)

[Abstract] The abstract refers to 'several mutation strategies' but does not enumerate or briefly characterize them; a short list or high-level taxonomy would improve readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for this observation on the abstract. We agree that the central empirical claims require more immediate support to allow verification, and we will revise the abstract accordingly while preserving its concise nature.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that 'our QDE variants achieve faster convergence and superior performance on several function classes in the BBOB benchmark' is asserted without any quantitative results, tables, statistical tests, error bars, or experimental-setup details. Because this empirical superiority is the load-bearing conclusion of the paper, the absence of supporting data prevents verification of the claim.

Authors: We acknowledge the validity of this point. The abstract currently summarizes the performance findings at a high level without quantitative anchors. In the revised version we will add a concise sentence reporting representative results (e.g., mean and standard deviation of final error on selected BBOB function groups, together with a brief statement of the experimental protocol: 20 independent runs, population size 100, maximum 10^5 evaluations). This will make the claim verifiable at a glance while directing readers to the full tables and statistical tests in Section 4. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on independent benchmark comparisons

full rationale

The paper introduces quaternion-valued DE mutation operators and reports faster convergence on BBOB functions relative to real-valued DE. No derivation step reduces to a self-definition, a fitted parameter relabeled as a prediction, or a load-bearing self-citation chain. The performance results are obtained from direct experimental runs on standard benchmarks, making the central claim externally falsifiable and independent of the algorithm's internal construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that quaternion algebra can be directly substituted into DE mutation without loss of algorithmic properties; no explicit free parameters or invented entities are introduced in the abstract.

axioms (1)

domain assumption Quaternion multiplication and addition can be used to define effective mutation operators that preserve or enhance DE's search behavior.
Invoked when the paper states that mutation strategies are designed to exploit algebraic and geometric properties of quaternions.

pith-pipeline@v0.9.0 · 5475 in / 1119 out tokens · 118527 ms · 2026-05-13T03:30:56.178977+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear
We propose six novel mutation strategies that leverage the algebraic and geometric properties of quaternions, including both Euclidean and polar-based mechanisms.
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
the use of quaternion algebra represents an emerging area in computational intelligence

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

[1]

Comple x- valued encoding metaheuristic optimization algorithm: A c omprehensive survey,

P . Wang, Y . Zhou, Q. Luo, C. Han, Y . Niu, and M. Lei, “Comple x- valued encoding metaheuristic optimization algorithm: A c omprehensive survey,” Neurocomputing, vol. 407, pp. 313–342, 2020

work page 2020
[2]

Modiﬁ ed ﬁreﬂy algorithm using quaternion representation,

I. Fister, X.-S. Y ang, J. Brest, and I. Fister Jr., “Modiﬁ ed ﬁreﬂy algorithm using quaternion representation,” Expert Systems with Applications , vol. 40, no. 18, pp. 7220–7230, 2013

work page 2013
[3]

Ki nematic synthesis of the passive human knee joint by differential ev olution and quaternions algebra: A preliminary study,

D. Saldias, L. Radavelli, C. Roesler, and D. Martins, “Ki nematic synthesis of the passive human knee joint by differential ev olution and quaternions algebra: A preliminary study,” in Proceedings of the 5th IEEE RAS/EMBS International Conference on Biomedical Robo tics and Biomechatronics, ser. BioRob, Sao Paulo, Brazil, 2014, pp. 96–101

work page 2014
[4]

Modiﬁed b at algorithm with quaternion representation,

I. Fister, J. Brest, I. Fister, and X.-S. Y ang, “Modiﬁed b at algorithm with quaternion representation,” in Proceedings of the IEEE Congress on Evolutionary Computation (CEC) , ser. CEC, Sendai, Japan, 2015, pp. 491–498

work page 2015
[5]

A genetic algorithm with multi-p arent crossover using quaternion representation for numerical f unction op- timization,

T. T. Khuat and M. H. Le, “A genetic algorithm with multi-p arent crossover using quaternion representation for numerical f unction op- timization,” Applied Intelligence, vol. 46, no. 4, p. 810–826, 2017

work page 2017
[6]

Protein– ligand docking using differential evolution with an adaptive mech anism,

S. Song, X. Chen, Y . Zhang, Z. Tang, and Y . Todo, “Protein– ligand docking using differential evolution with an adaptive mech anism,” Knowledge-Based Systems , vol. 231, p. 107433, 2021

work page 2021
[7]

Complex and hypercomplex-valued sup port vector machines: A survey,

N. Arana-Daniel, “Complex and hypercomplex-valued sup port vector machines: A survey,” Applied Sciences , vol. 9, no. 15, 2019

work page 2019
[8]

A survey of qu aternion neural networks,

T. Parcollet, M. Morchid, and G. Lin` ares, “A survey of qu aternion neural networks,” Artiﬁcial Intelligence Review , vol. 53, no. 4, pp. 2957–2982, 2020

work page 2020
[9]

Quaternion con volutional neural networks: Current advances and future directions,

G. Altamirano-Gomez and C. Gershenson, “Quaternion con volutional neural networks: Current advances and future directions,” Advances in Applied Clifford Algebras , vol. 34, no. 42, pp. 1–63, 2024

work page 2024
[10]

Simultaneously learning deep quater- nion reconstruction and noise convolutional dictionary fo r color image denoising,

Z. Zhou, Y . Chen, and Y . Zhou, “Simultaneously learning deep quater- nion reconstruction and noise convolutional dictionary fo r color image denoising,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 9, no. 2, pp. 1766–1779, 2025

work page 2025
[11]

Quaternion capsule neu ral network with region attention for facial expression recognition in color images,

Y . Zhou, L. Jin, G. Ma, and X. Xu, “Quaternion capsule neu ral network with region attention for facial expression recognition in color images,” IEEE Transactions on Emerging Topics in Computational Inte lligence, vol. 6, no. 4, pp. 893–912, 2022

work page 2022
[12]

A quaternion-valued ne ural network approach to nonsmooth nonconvex constrained optimization in quater- nion domain,

J. Liu, X. Liao, and J.-S. Dong, “A quaternion-valued ne ural network approach to nonsmooth nonconvex constrained optimization in quater- nion domain,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 1, pp. 654–669, 2024

work page 2024
[13]

Quater nion dendritic neuron model for multivariate ﬁnancial time seri es prediction,

Q. Y u, Z. Zhang, Z. Wang, H. Li, Z. Lei, and S. Gao, “Quater nion dendritic neuron model for multivariate ﬁnancial time seri es prediction,” IEEE Transactions on Emerging Topics in Computational Inte lligence, pp. 1–12, 2024

work page 2024
[14]

Keyword spotting with quaternionic ResNet: Application to spottin g in Greek manuscripts,

G. Sﬁkas, G. Retsinas, A. Giotis, B. Gatos, and C. Nikou, “Keyword spotting with quaternionic ResNet: Application to spottin g in Greek manuscripts,” in Document Analysis Systems: 15th IAPR International W orkshop, ser. DAS, La Rochelle, France, 2022, pp. 382–396

work page 2022
[15]

Hamilton, Lectures on quaternions: Containing a systematic state- ment of a new mathematical method

W. Hamilton, Lectures on quaternions: Containing a systematic state- ment of a new mathematical method . Dublin, England: Hodges and Smith, Whittaker & Co., MacMillan & Co., 1853

work page
[16]

London, England: Longmans, Green, & Co., 1866

——, Elements of quaternions . London, England: Longmans, Green, & Co., 1866

work page
[17]

On quaternions, or on a new system of imaginaries in algebra,

W. R. Hamilton, “On quaternions, or on a new system of imaginaries in algebra,” 2000, https://www.maths.tcd.ie/pub/HistMath/People/Hamilton/OnQuat/

work page 2000
[18]

Kantor and A

I. Kantor and A. Solodovnikov, Hypercomplex numbers . New Y ork, NY: Springer-V erlag, 1989

work page 1989
[19]

Ward, Quaternions and Cayley numbers

J. Ward, Quaternions and Cayley numbers . Dordrecht, Netherlands: Kluwer Academic Publishers, 1997

work page 1997
[20]

Hanson, Visualizing quaternions

A. Hanson, Visualizing quaternions. San Francisco, CA: Elsevier Inc., 2006

work page 2006
[21]

Du V al, Homographies, quaternions and rotations

P . Du V al, Homographies, quaternions and rotations. London, England: Oxford University Press, 1964

work page 1964
[22]

Co mparing results of 31 algorithms from the black-box optimization be nchmarking bbob-2009,

N. Hansen, A. Auger, R. Ros, S. Finck, and P . Poˇ s´ ık, “Co mparing results of 31 algorithms from the black-box optimization be nchmarking bbob-2009,” in Proceedings of the 12th Annual Conference Companion on Genetic and Evolutionary Computation , ser. GECCO ’10, Portland, Oregon, USA, 2010, pp. 1689–1696

work page 2009
[23]

Statistical comparisons of classiﬁers ov er multiple data sets,

J. Demˇ sar, “Statistical comparisons of classiﬁers ov er multiple data sets,” Journal of Machine Learning Research , vol. 7, pp. 1–30, 2006

work page 2006
[24]

No free lunch theorems for s earch,

D. Wolpert and W. Macready, “No free lunch theorems for s earch,” The Santa Fe Institute, Santa Fe, NM, Tech. Rep. SFI-WP-95-02- 0 10, Feb. 1995

work page 1995
[25]

No free lunch theorems for optimization,

D. H. Wolpert and W. G. Macready, “No free lunch theorems for optimization,” IEEE Transactions on Evolutionary Computation , vol. 1, no. 1, pp. 67–82, 1997

work page 1997